| ClusterTest-class {ClassDiscovery} | R Documentation |
This is a base class for tests that attempt to determine whether the groups found by an unsupervised clustering method are statistically significant.
## S4 method for signature 'ClusterTest': image(x, dendrogram, ...)
x |
An object of the ClusterTest class. |
dendrogram |
An object with S3 class hclust, as returned
by the hclust function. |
... |
Additional graphical parameters to be passed to the
standard image function. |
Objects can be created by calls of the form new("ClusterTest", ...).
Most users, however, will only create objects from one of the derived
classes such as BootstrapClusterTest or
PerturbationClusterTest.
call:call, which shows
how the object was constructed.result:matrix containing the
results of the cluster reproducibility test. The size of the
matrix corresponds to the number of samples (columns) in the data
set on which the test was performed. The result matrix
should contain "agreement" values between 0 and 1, representing for
each pair of samples the fraction of times that they were
collected into the same cluster.signature(x = "ClusterTest"): Produces a
histogram of the agreement fractions. When a true group structure
exists, one expects a multimodal distribution,with one group of
agreements near 0 (for pairs belonging to different clusters) and
one group of agreements near 1 (for pairs belonging to the same
cluster).signature(x = "ClusterTest"): Uses the
heatmap function to display the agreement matrix. The
optional dendrogram argument should be used to display the
extent to which the ageement matrix matches the results of
hierarchical clustering using the full data set.signature(object = "ClusterTest"): Write out a
summary of the object.Kevin R. Coombes <kcoombes@mdanderson.org>
Kerr MK, Churchill GJ. Boostrapping cluster analysis: Assessing the reliability of conclusions from microarray experiments. PNAS 2001; 98:8961-8965.
BootstrapClusterTest,
PerturbationClusterTest, heatmap
# simulate data from two different classes
d1 <- matrix(rnorm(100*30, rnorm(100, 0.5)), nrow=100, ncol=30, byrow=FALSE)
d2 <- matrix(rnorm(100*20, rnorm(100, 0.5)), nrow=100, ncol=20, byrow=FALSE)
dd <- cbind(d1, d2)
# cluster the data
hc <- hclust(distanceMatrix(dd, 'pearson'), method='average')
# make a fake reproducibility matrix
fraud <- function(x) {
new('ClusterTest', result=abs(cor(x)), call=match.call())
}
fake <- fraud(dd)
summary(fake)
hist(fake)
image(fake) # let heatmap compute a new dendrogram from the agreements
image(fake, dendrogram=hc) # use the actual dendrogram from the data
image(fake, dendrogram=hc, col=blueyellow(64)) # change the colors
#cleanup
rm(fake, fraud, hc, dd, d1, d2)